compiler: add explicit convert in Type_guard_expression::do_get_backend

    The current recipe in Type_guard_expression for creating a Bexpression
    performs a type conversion at the Type level, but doesn't invoke
    Backend::convert_expression to capture the conversion for the back
    end. Add code to create a BE type conversion operation.

    Reviewed-on: https://go-review.googlesource.com/50373

From-SVN: r250400

re PR target/80846 (auto-vectorized AVX2 horizontal sum should narrow to 128b right away, to be more efficient for Ryzen and Intel)

PR target/80846
* config/i386/i386.c (ix86_expand_vector_init_general): Handle
V2TImode and V4TImode.
(ix86_expand_vector_extract): Likewise.
* config/i386/sse.md (VMOVE): Enable V4TImode even for just
TARGET_AVX512F, instead of only for TARGET_AVX512BW.
(ssescalarmode): Handle V4TImode and V2TImode.
(VEC_EXTRACT_MODE): Add V4TImode and V2TImode.
(*vec_extractv2ti, *vec_extractv4ti): New insns.
(VEXTRACTI128_MODE): New mode iterator.
(splitter for *vec_extractv?ti first element): New.
(VEC_INIT_MODE): New mode iterator.
(vec_init<mode>): Consolidate 3 expanders into one using
VEC_INIT_MODE mode iterator.

* gcc.target/i386/avx-pr80846.c: New test.
* gcc.target/i386/avx2-pr80846.c: New test.
* gcc.target/i386/avx512f-pr80846.c: New test.

From-SVN: r250397

lra-assigns.c: fix pseudo_compare_func

* lra-assigns.c (pseudo_compare_func): Fix comparison step based on
non_spilled_static_chain_regno_p.

From-SVN: r250395

gimple-ssa-store-merging.c: fix sort_by_bitpos

* gimple-ssa-store-merging.c (sort_by_bitpos): Return 0 on equal bitpos.

From-SVN: r250394

bb-reorder.c (connect_traces): Allow copying of blocks within single partition.

* bb-reorder.c (connect_traces): Allow copying of blocks within
single partition.

From-SVN: r250390

Minor tweaks

From-SVN: r250386

gimple.h (gimple_phi_result): Add gphi * overload.

2017-07-20  Richard Biener  <rguenther@suse.de>

* gimple.h (gimple_phi_result): Add gphi * overload.
(gimple_phi_result_ptr): Likewise.
(gimple_phi_arg): Likewise.  Adjust index assert to only
allow actual argument accesses rather than all slots available
by capacity.
(gimple_phi_arg_def): Add gphi * overload.
* tree-phinodes.c (make_phi_node): Initialize only actual
arguments.
(resize_phi_node): Clear memory not covered by old node,
do not initialize excess argument slots.
(reserve_phi_args_for_new_edge): Initialize new argument slot
completely.

From-SVN: r250385

re PR tree-optimization/81388 (Incorrect code generation with -O1)

PR tree-optimization/81388
Revert r238585:
2016-07-21  Bin Cheng  <bin.cheng@arm.com>

* tree-ssa-loop-niter.c (number_of_iterations_lt_to_ne): Clean up
by removing computation of may_be_zero.

gcc/testsuite
PR tree-optimization/81388
* gcc.dg/tree-ssa/pr81388-1.c: New test.
* gcc.dg/tree-ssa/pr81388-2.c: New test.

From-SVN: r250384

re PR middle-end/81030 (ICE on valid code at -O1 (only) on x86_64-linux-gnu: verify_flow_info failed)

PR middle-end/81030
* cfgbuild.c (find_many_sub_basic_blocks): Update REG_BR_PROB note
when gimple level profile disagrees with what RTL expander did.
* gcc.dg/pr81030.c: New test.

From-SVN: r250383

re PR tree-optimization/61171 (vectorization fails for a reduction in presence of subtraction)

2017-07-20  Richard Biener  <rguenther@suse.de>

PR tree-optimization/61171
* tree-vectorizer.h (slp_instance): Add reduc_phis member.
(vect_analyze_stmt): Add slp instance parameter.
(vectorizable_reduction): Likewise.
* tree-vect-loop.c (vect_analyze_loop_operations): Adjust.
(vect_is_simple_reduction): Deal with chains not detected
as SLP reduction chain, specifically not properly associated
chains containing a mix of plus/minus.
(get_reduction_op): Remove.
(get_initial_defs_for_reduction): Simplify, pass in whether
this is a reduction chain, pass in the SLP node for the PHIs.
(vect_create_epilog_for_reduction): Get the SLP instance as
arg and adjust.
(vectorizable_reduction): Get the SLP instance as arg.
During analysis remember the SLP node with the PHIs in the
instance.  Simplify getting at the vectorized reduction PHIs.
* tree-vect-slp.c (vect_slp_analyze_node_operations): Pass
through SLP instance.
(vect_slp_analyze_operations): Likewise.
* tree-vect-stms.c (vect_analyze_stmt): Likewise.
(vect_transform_stmt): Likewise.

* g++.dg/vect/pr61171.cc: New testcase.
* gfortran.dg/vect/pr61171.f: Likewise.
* gcc.dg/vect/vect-reduc-11.c: Likewise.

From-SVN: r250382

slp-43.c: Increase loop count to enable vectorization with V64QImode.

2017-07-20  Richard Biener  <rguenther@suse.de>

* gcc.dg/vect/slp-43.c: Increase loop count to enable vectorization
with V64QImode.
* gcc.dg/vect/slp-45.c: Likewise.

From-SVN: r250380

Fix phi arg location in find_implicit_erroneous_behavior

2017-07-20  Tom de Vries  <tom@codesourcery.com>

PR tree-optimization/81489
* gimple-ssa-isolate-paths.c (find_implicit_erroneous_behavior): Move
read of phi arg location to before loop that modifies phi.

From-SVN: r250378

match.pd (((m1 >/</>=/<= m2) * d -> (m1 >/</>=/<= m2) ? d : 0): New pattern.

gcc
* match.pd (((m1 >/</>=/<= m2) * d -> (m1 >/</>=/<= m2) ? d : 0):
New pattern.

gcc/testsuite
* gcc.dg/tree-ssa/vrp116.c: New Test.

From-SVN: r250377

Daily bump.

From-SVN: r250376

re PR regression/81331 (missed Eh delivery in partitioned function)

PR middle-end/81331
* except.c (execute): Fix ordering issue.

From-SVN: r250370

* math/powq.c (powq): Use uint32_t instead of u_int32_t.

From-SVN: r250367

PR libstdc++/81476 Optimise vector insertion from input iterators

PR libstdc++/81476
* include/bits/vector.tcc (vector::_M_range_insert<_InputIterator>):
Only insert elements one-by-one when inserting at the end.
* testsuite/performance/23_containers/insert/81476.cc: New.

From-SVN: r250366

combine: Fix for PR81423

We here have an AND of a SUBREG of an LSHIFTRT.  If that SUBREG is
paradoxical, the extraction we form is the length of the size of the
inner mode, which includes some bits that should not be in the result.
Just give up in that case.

PR rtl-optimization/81423
* combine.c (make_compound_operation_int): Don't try to optimize
the AND of a SUBREG of an LSHIFTRT if that SUBREG is paradoxical.

From-SVN: r250365

cpu-builtin-1.c: Change test to use #ifdef __BUILTIN_CPU_SUPPORTS to see if...

2017-07-19  Michael Meissner  <meissner@linux.vnet.ibm.com>

* gcc.target/powerpc/cpu-builtin-1.c: Change test to use #ifdef
__BUILTIN_CPU_SUPPORTS to see if the GLIBC is new enough that
__builtin_cpu_is and __builtin_cpu_supports are supported.

From-SVN: r250364

simplify-rtx: The truncation of an IOR can have all bits set (PR81423)

... if it is an IOR with a constant with all bits set in the mode
that is truncated to, for example.  Handle that case.

PR rtl-optimization/81423
* simplify-rtx.c (simplify_truncation): Handle truncating an IOR
with a constant that is -1 in the truncated to mode.

From-SVN: r250363

Fix up plafform testes in check headers.

Fix up plafform testes in check headers. After a resent GCC change
the previously submitted BMI/BMI2 intrinsic test started to fail
with a warning/error.

[gcc/testsuite]

2017-07-19  Steven Munroe  <munroesj@gcc.gnu.org>

* gcc.target/powerpc/bmi-check.h (main): Skip unless
__BUILTIN_CPU_SUPPORTS__ defined.
* gcc.target/powerpc/bmi2-check.h (main): Skip unless
__BUILTIN_CPU_SUPPORTS__ defined.

From-SVN: r250362

predict.c (propagate_unlikely_bbs_forward): Break out from ...

* predict.c (propagate_unlikely_bbs_forward): Break out from ...
(determine_unlikely_bbs): ... here.
* predict.h (propagate_unlikely_bbs_forward): Declare.
* cfgexpand.c (pass_expand::execute): Use it.
* bb-reorder.c (sanitize_hot_paths): Do not consider known to be
unlikely edges.
(find_rarely_executed_basic_blocks_and_crossing_edges): Use
propagate_unlikely_bbs_forward.

From-SVN: r250360

predict.c (propagate_unlikely_bbs_forward): Break out from ...

* predict.c (propagate_unlikely_bbs_forward): Break out from ...
(determine_unlikely_bbs): ... here.
* predict.h (propagate_unlikely_bbs_forward): Declare.
* cfgexpand.c (pass_expand::execute): Use it.
* bb-reorder.c (sanitize_hot_paths): Do not consider known to be
unlikely edges.
(find_rarely_executed_basic_blocks_and_crossing_edges): Use
propagate_unlikely_bbs_forward.

From-SVN: r250359

re PR regression/81331 (missed Eh delivery in partitioned function)

PR middle-end/81331
* except.c (maybe_add_nop_after_section_switch): New function.
(execute): Use it.

From-SVN: r250358

new (bad_array_new_length): Remove redundant semicolon after in-class member function definition.

        * libsupc++/new (bad_array_new_length): Remove redundant
        semicolon after in-class member function definition.
        * include/bits/locale_facets.h (ctype_byname, num_put): Likewise.
        * include/bits/locale_facets_nonio.h (time_put_byname): Likewise.
        * include/bits/random.h (mersenne_twister_engine): Likewise.
        * include/tr1/random.h (mersenne_twister): Likewise
        * include/ext/random (simd_fast_mersenne_twister_engine): Likewise.
        * include/ext/rope (char_producer, _Rope_char_consumer,
        _Rope_self_destruct_ptr, _Rope_const_iterator, _Rope_iterator):
        Likewise.
        * include/ext/ropeimpl.h (_Rope_flatten_char_consumer,
        _Rope_insert_char_consumer): Likewise.

From-SVN: r250357

ada-tree.h (TYPE_OBJECT_RECORD_TYPE, [...]): Use TYPE_MIN_VALUE_RAW instead of TYPE_MINVAL.

* gcc-interface/ada-tree.h (TYPE_OBJECT_RECORD_TYPE,
TYPE_GCC_MIN_VALUE): Use TYPE_MIN_VALUE_RAW instead of TYPE_MINVAL.
(TYPE_GCC_MAX_VALUE): Use TYPE_MAX_VALUE_RAW instead of TYPE_MAXVAL.

From-SVN: r250355

Fix Unwind support on DragonFly BSD after sigtramp move

2017-07-19  John Marino  <gnugcc@marino.st>

* config/i386/dragonfly-unwind.h: Handle sigtramp relocation.

From-SVN: r250354

Fix indentation in std::basic_filebuf member functions

* include/bits/fstream.tcc (basic_filebuf::xsgetn)
(basic_filebuf::xsputn, basic_filebuf::seekoff): Fix indentation.

From-SVN: r250353

Fix assert in gimple_phi_set_arg

2017-07-19  Tom de Vries  <tom@codesourcery.com>

* gimple.h (gimple_phi_set_arg): Make assert more strict.

From-SVN: r250352

Fix assert in gimple_phi_arg

2017-07-19  Tom de Vries  <tom@codesourcery.com>

* gimple.h (gimple_phi_arg): Make assert more strict.

From-SVN: r250351

Now adding the mmintrin.h intrinsic headers. The DG tests will follow.

2017-07-19  Steven Munroe  <munroesj@gcc.gnu.org>

* config.gcc (powerpc*-*-*): Add mmintrin.h.
* config/rs6000/mmintrin.h: New file.
* config/rs6000/x86intrin.h [__ALTIVEC__]: Include mmintrin.h.

From-SVN: r250350

Fix indentation in libstdc++ testsuite utils

* testsuite/util/testsuite_iterators.h: Fix indentation.

From-SVN: r250349

class.c (add_implicitly_declared_members): Use classtype_has_move_assign_or_move_ctor_p.

* class.c (add_implicitly_declared_members): Use
classtype_has_move_assign_or_move_ctor_p.
(classtype_has_move_assign_or_move_ctor,
classtype_has_user_move_assign_or_move_ctor_p): Merge into ...
(classtype_has_move_assign_or_move_ctor_p): ... this new function.
* cp-tree.h (classtype_has_user_move_assign_or_move_ctor_p):
Replace with ...
(classtype_has_move_assign_or_move_ctor_p): ... this.
* method.c (maybe_explain_implicit_delete, lazily_declare_fn): Adjust.
* tree.c (type_has_nontrivial_copy_init): Adjust.

From-SVN: r250344

re PR libquadmath/65757 (gfortran gives incorrect result for anint with real*16 argument)

PR libquadmath/65757
* quadmath-imp.h (math_opt_barrier, math_force_eval,
math_narrow_eval, math_check_force_underflow,
math_check_force_underflow_nonneg): Define.
* math/ceilq.c: Backport changes from upstream glibc
between 2012-11-01 and 2017-07-13.
* math/remquoq.c: Likewise.
* math/expq.c: Likewise.
* math/llroundq.c: Likewise.
* math/logq.c: Likewise.
* math/atanq.c: Likewise.
* math/nearbyintq.c: Likewise.
* math/scalblnq.c: Likewise.
* math/finiteq.c: Likewise.
* math/atanhq.c: Likewise.
* math/expm1q.c: Likewise.
* math/sinhq.c: Likewise.
* math/log10q.c: Likewise.
* math/rintq.c: Likewise.
* math/roundq.c: Likewise.
* math/fmaq.c: Likewise.
* math/erfq.c: Likewise.
* math/log2q.c: Likewise.
* math/lroundq.c: Likewise.
* math/j1q.c: Likewise.
* math/scalbnq.c: Likewise.
* math/truncq.c: Likewise.
* math/frexpq.c: Likewise.
* math/sincosq.c: Likewise.
* math/tanhq.c: Likewise.
* math/asinq.c: Likewise.
* math/coshq.c: Likewise.
* math/j0q.c: Likewise.
* math/asinhq.c: Likewise.
* math/floorq.c: Likewise.
* math/sinq_kernel.c: Likewise.
* math/powq.c: Likewise.
* math/hypotq.c: Likewise.
* math/sincos_table.c: Likewise.
* math/rem_pio2q.c: Likewise.
* math/nextafterq.c: Likewise.
* math/log1pq.c: Likewise.
* math/sincosq_kernel.c: Likewise.
* math/tanq.c: Likewise.
* math/acosq.c: Likewise.
* math/lrintq.c: Likewise.
* math/llrintq.c: Likewise.

From-SVN: r250343

re PR tree-optimization/81346 (Missed constant propagation into comparison)

PR tree-optimization/81346
* match.pd: Optimize (X - 1U) <= INT_MAX-1U into (int) X > 0.

* gcc.dg/tree-ssa/pr81346-5.c: New test.

From-SVN: r250342

Add v2di support for nvptx

2017-07-19  Tom de Vries  <tom@codesourcery.com>

* config/nvptx/nvptx-modes.def: Add V2DImode.
* config/nvptx/nvptx-protos.h (nvptx_data_alignment): Declare.
* config/nvptx/nvptx.c (nvptx_ptx_type_from_mode): Handle V2DImode.
(nvptx_output_mov_insn): Handle lack of mov.b128.
(nvptx_print_operand): Handle 'H' and 'L' codes.
(nvptx_vector_mode_supported): Allow V2DImode.
(nvptx_preferred_simd_mode): New function.
(nvptx_data_alignment): New function.
(TARGET_VECTORIZE_PREFERRED_SIMD_MODE): Redefine to
nvptx_preferred_simd_mode.
* config/nvptx/nvptx.h (STACK_BOUNDARY, BIGGEST_ALIGNMENT): Change from
64 to 128 bits.
(DATA_ALIGNMENT): Define.  Set to nvptx_data_alignment.

* config/nvptx/nvptx.md (VECIM): Add V2DI.

* gcc.target/nvptx/decl-init.c: Update alignment.
* gcc.target/nvptx/slp-2-run.c: New test.
* gcc.target/nvptx/slp-2.c: New test.
* gcc.target/nvptx/v2di.c: New test.

* testsuite/libgomp.oacc-c/vec.c: New test.

From-SVN: r250341

Add v2si support for nvptx

2017-07-19  Tom de Vries  <tom@codesourcery.com>

* config/nvptx/nvptx-modes.def: New file.  Add V2SImode.
* config/nvptx/nvptx.c (nvptx_ptx_type_from_mode): Handle V2SImode.
(nvptx_vector_mode_supported): New function.  Allow V2SImode.
(TARGET_VECTOR_MODE_SUPPORTED_P): Redefine to nvptx_vector_mode_supported.
* config/nvptx/nvptx.md (VECIM): New mode iterator. Add V2SI.
(mov<VECIM>_insn): New define_insn.
(define_expand "mov<VECIM>): New define_expand.

* gcc.target/nvptx/slp-run.c: New test.
* gcc.target/nvptx/slp.c: New test.
* gcc.target/nvptx/v2si-cvt.c: New test.
* gcc.target/nvptx/v2si-run.c: New test.
* gcc.target/nvptx/v2si.c: New test.
* gcc.target/nvptx/vec.inc: New test.

From-SVN: r250340

Add generic v2 vector mode support for nvptx

2017-07-19  Tom de Vries  <tom@codesourcery.com>

* config/nvptx/nvptx.c (nvptx_print_operand): Handle v2 vector mode.

From-SVN: r250339

re PR tree-optimization/81346 (Missed constant propagation into comparison)

PR tree-optimization/81346
* fold-const.h (fold_div_compare, range_check_type): Declare.
* fold-const.c (range_check_type): New function.
(build_range_check): Use range_check_type.
(fold_div_compare): No longer static, rewritten into
a match.pd helper function.
(fold_comparison): Don't call fold_div_compare here.
* match.pd (X / C1 op C2): New optimization using fold_div_compare
as helper function.

* gcc.dg/tree-ssa/pr81346-1.c: New test.
* gcc.dg/tree-ssa/pr81346-2.c: New test.
* gcc.dg/tree-ssa/pr81346-3.c: New test.
* gcc.dg/tree-ssa/pr81346-4.c: New test.
* gcc.target/i386/umod-3.c: Hide comparison against 1 from the
compiler to avoid X / C1 op C2 optimization to trigger.

From-SVN: r250338

tree.h (TYPE_MINVAL, [...]): Rename to ...

gcc/
* tree.h (TYPE_MINVAL, TYPE_MAXVAL): Rename to ...
(TYPE_MIN_VALUE_RAW, TYPE_MAX_VALUE_RAW): ... these.
* tree.c (find_decls_types_r, verify_type): Use
TYPE_{MIN,MAX}_VALUE_RAW.
* lto-streamer-out.c (DFS::DFS_write_tree_body): Likewise.
(hash_tree): Likewise.
* tree-streamer-in.c (lto_input_ts_type_non_common_tree_pointers):
Likewise.
* tree-streamer-out.c (write_ts_type_non_common_tree_pointers):
Likewise.

gcc/cp/
* cp-tree.h (PACK_EXPANSION_PARAMETER_PACKS,
PACK_EXPANSION_EXTRA_ARGS): Use TYPE_{MIN,MAX}_VALUE_RAW.

gcc/lto/
* lto.c (mentions_vars_p_type): Use TYPE_{MIN,MAX}_VALUE_RAW.
(compare_tree_sccs_1, lto_fixup_prevailing_decls): Likewise.

gcc/objc/
* objc-act.h (CLASS_NST_METHODS, CLASS_CLS_METHODS): Use
TYPE_{MIN,MAX}_VALUE_RAW.

From-SVN: r250337

Fix an UBSAN test-case (PR sanitizer/63361).

2017-07-19  Martin Liska  <mliska@suse.cz>

PR sanitizer/63361
* c-c++-common/ubsan/float-cast-overflow-1.c: Add either
-ffloat-store or -mieee for targets that need it.

From-SVN: r250336

Handle equal-argument loop exit phi in expand_omp_for_static_chunk

2017-07-18  Tom de Vries  <tom@codesourcery.com>

PR middle-end/81464
* omp-expand.c (expand_omp_for_static_chunk): Handle equal-argument loop
exit phi.

* gfortran.dg/pr81464.f90: New test.

From-SVN: r250335

Add myself to MAINTAINERS.

From-SVN: r250334

Use dg-require-fileio in new test

* testsuite/27_io/basic_filebuf/sgetn/char/81395.cc: Add dg-require.

From-SVN: r250333

Daily bump.

From-SVN: r250332

PR libstdc++/81395 fix crash when write follows large read

PR libstdc++/81395
* include/bits/fstream.tcc (basic_filebuf::xsgetn): Don't set buffer
pointers for write mode after reading.
* testsuite/27_io/basic_filebuf/sgetn/char/81395.cc: New.

From-SVN: r250328

compiler: insert backend type conversion for closure func ptr

    In Func_expression::do_get_backend when creating the backend
    representation for a closure, create a backend type conversion to
    account for potential differences between the closure struct type
    (where the number of fields is dependent on the number of values
    referenced in the closure) and the generic function descriptor type
    (struct with single function pointer field).

    Reviewed-on: https://go-review.googlesource.com/49255

From-SVN: r250327

re PR go/81451 (missing futex check - libgo/runtime/thread-linux.c:12:0 futex.h:13:12: error: expected ‘=’, ‘,’, ‘;’, ‘asm’ or ‘__attribute__’ before ‘long’)

PR go/81451
    runtime: inline runtime_osinit

    We had two identical copies of runtime_osinit. They set runtime_ncpu,
    a variable that is no longer used. Removing that leaves us with two lines.
    Inline those two lines in the two places the function was called.

    This fixes GCC PR 81451.

    Reviewed-on: https://go-review.googlesource.com/48862

From-SVN: r250326

compiler: pass correct 'function' flag to circular_pointer_type

    The code in Named_type::do_get_backend was not passing the correct
    flag value for circular function types to Backend::circular_pointer_type
    (it was always setting this flag to false). Pass a true value if the
    type being converted is a function type.

    Reviewed-on: https://go-review.googlesource.com/49330

From-SVN: r250325

re PR go/81324 (libgo does not build with glibc 2.18)

PR go/81324
    sysinfo.c: ignore ptrace_peeksiginfo_args from <linux/ptrace.h>

    With some versions of glibc and GNU/Linux ptrace_pseeksiginfo_args is
    defined in both <sys/ptrace.h> and <linux/ptrace.h>. We don't actually
    care about the struct, so use a #define to avoid a redefinition error.

    This fixes https://gcc.gnu.org/PR81324.

    Reviewed-on: https://go-review.googlesource.com/49290

From-SVN: r250324

update my entry in MAINTAINERS

/:

* MAINTAINERS: Update my email address.

From-SVN: r250321

stl_list.h (struct _List_node_header): New.

2017-07-18  François Dumont  <fdumont@gcc.gnu.org>

* include/bits/stl_list.h
(struct _List_node_header): New.
(_List_impl()): Fix noexcept qualification.
(_List_impl(_List_impl&&)): New, default.
(_List_base()): Default.
(_List_base(_List_base&&)): Default.
(_List_base::_M_move_nodes): Adapt to use
_List_node_header._M_move_nodes.
(_List_base::_M_init): Likewise.
(list<>()): Default.
(list<>(list&&)): Default.
(list<>::_M_move_assign(list&&, true_type)): Use _M_move_nodes.
(__distance(_List_const_iterator<>, _List_const_iterator<>,
input_iterator_tag)): Adapt.
* testsuite/23_containers/list/allocator/default_init.cc: New.

From-SVN: r250320

re PR target/81471 (internal compiler error: in curr_insn_transform, at lra-constraints.c:3495)

PR target/81471
* config/i386/i386.md (rorx_immediate_operand): New mode attribute.
(*bmi2_rorx<mode>3_1): Use rorx_immediate_operand as
operand 2 predicate.
(*bmi2_rorxsi3_1_zext): Use const_0_to_31_operand as
operand 2 predicate.
(ror,rol -> rorx splitters): Use const_int_operand as
operand 2 predicate.

testsuite/ChangeLog:

PR target/81471
* gcc.target/i386/pr81471.c: New test.

From-SVN: r250315

mklog: Fix extraction of changed file name.

2017-07-17  Yury Gribov  <tetra2005@gmail.com>

* mklog: Fix extraction of changed file name.

From-SVN: r250314

re PR tree-optimization/81410 (-O3 breaks code)

2017-06-18  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81410
* tree-vect-stmts.c (vectorizable_load): Properly adjust for
the gap in the ! slp_perm SLP case after each group.

* gcc.dg/vect/pr81410.c: New testcase.

From-SVN: r250312

re PR tree-optimization/81463 (ICE in scale_loop_profile at gcc/cfgloopmanip.c:603)

PR middle-end/81463
* cfgloopmanip.c (scale_loop_profile): Watch out for zero frequency
again.

From-SVN: r250311

re PR tree-optimization/81462 (ICE in estimate_bb_frequencies at gcc/predict.c:3546)

PR middle-end/81462
* predict.c (set_even_probabilities): Cleanup; do not affect
probabilities that are already known.
(combine_predictions_for_bb): Call even when count is set.

* g++.dg/torture/pr81462.C: New testcase.

From-SVN: r250310

tree-parloops.c (try_transform_to_exit_first_loop_alt): Use TYPE_MAX_VALUE.

       gcc/
* tree-parloops.c (try_transform_to_exit_first_loop_alt): Use
TYPE_MAX_VALUE.

       gcc/c-family/
* c-warn.c (warn_for_memset): Use TYPE_{MIN,MAX}_VALUE.

       gcc/c/
* c-parser.c (c_parser_array_notation): Use TYPE_{MIN,MAX}_VALUE.

       gcc/cp/
* cp-array-notation.c (build_array_notation_ref): Use
TYPE_{MIN,MAX}_VALUE.

       gcc/fortran/
* trans.c (gfc_build_array_ref): Use TYPE_MAX_VALUE.

From-SVN: r250309

gcc: ada: delete old $(P) reference

The P variable was deleted back in Nov 2015 (svn rev 231062),
but its expansion was missed.  Delete those now too.

From-SVN: r250307

class.c (classtype_has_move_assign_or_move_ctor): Declare.

* class.c (classtype_has_move_assign_or_move_ctor): Declare.
(add_implicitly_declared_members): Use it.
(type_has_move_constructor, type_has_move_assign): Merge into ...
(classtype_has_move_assign_or_move_ctor): ... this new function.
* cp-tree.h (type_has_move_constructor, type_has_move_assign): Delete.

From-SVN: r250305

re PR middle-end/81408 (Lots of new -Wunsafe-loop-optimizations warnings with 7 compared to 6)

PR target/81408
* tree-ssa-loop-niter.c (number_of_iterations_exit): Dump missed
optimization for loop niter analysis.

gcc/testsuite
* g++.dg/tree-ssa/pr81408.C: New.
* gcc.dg/tree-ssa/pr19210-1.c: Check dump message rather than warning.

From-SVN: r250304

re PR target/81473 ([avr] build fails due to INT8_MIN and friends.)

gcc/
PR target/81473
* config/avr/avr.c (avr_optimize_casesi): Don't use
INT8_MIN, INT8_MAX, UINT8_MAX, INT16_MIN, INT16_MAX, UINT16_MAX.

From-SVN: r250301

Fix PR81362: Vector peeling

npeel was erroneously overwritten by vect_peeling_hash_get_lowest_cost
although the corresponding dataref is not used afterwards.  It should
be safe to get rid of the npeel parameter since we use the returned
peeling_info's npeel anyway.  Also removed the body_cost_vec parameter
which is not used elsewhere.

gcc/ChangeLog:

2017-07-18  Robin Dapp  <rdapp@linux.vnet.ibm.com>

* tree-vect-data-refs.c (vect_enhance_data_refs_alignment): Remove
body_cost_vec from _vect_peel_extended_info.
(vect_peeling_hash_get_lowest_cost): Do not set body_cost_vec.
(vect_peeling_hash_choose_best_peeling): Remove body_cost_vec and
npeel.

From-SVN: r250300

* config/arm/arm.c (emit_unlikely_jump): Remove unused var.

From-SVN: r250299

re PR tree-optimization/80620 (gcc produces wrong code with -O3)

2017-07-18  Richard Biener  <rguenther@suse.de>

PR tree-optimization/80620
PR tree-optimization/81403
* tree-ssa-pre.c (phi_translate_1): Clear range and points-to
info when re-using a VN table entry.

* gcc.dg/torture/pr80620.c: New testcase.
* gcc.dg/torture/pr81403.c: Likewise.

From-SVN: r250297

re PR tree-optimization/81418 (ICE in vect_get_vec_def_for_stmt_copy)

2017-06-18  Richard Biener  <rguenther@suse.de>

PR tree-optimization/81418
* tree-vect-loop.c (vectorizable_reduction): Properly compute
vectype_in.  Verify that with lane-reducing reduction operations
we have a single def-use cycle.

* gcc.dg/torture/pr81418.c: New testcase.

From-SVN: r250296

Revert commit r249424 2017-06-20 Carl Love <cel@us.ibm.com>

gcc/ChangeLog:

2017-07-17  Carl Love  <cel@us.ibm.com>

Revert commit r249424 2017-06-20  Carl Love  <cel@us.ibm.com>

* config/rs6000/rs6000-c.c (altivec_overloaded_builtins): Add
ALTIVEC_BUILTIN_VMULESW, ALTIVEC_BUILTIN_VMULEUW,
ALTIVEC_BUILTIN_VMULOSW, ALTIVEC_BUILTIN_VMULOUW entries.
* config/rs6000/rs6000.c (rs6000_gimple_fold_builtin,
builtin_function_type): Add ALTIVEC_BUILTIN_* case statements.
* config/rs6000/altivec.md (MVULEUW, VMULESW, VMULOUW,
VMULOSW): New enum "unspec" values.
(vec_widen_umult_even_v4si, vec_widen_smult_even_v4si,
vec_widen_umult_odd_v4si, vec_widen_smult_odd_v4si,
altivec_vmuleuw, altivec_vmulesw, altivec_vmulouw,
altivec_vmulosw): New patterns.
* config/rs6000/rs6000-builtin.def (VMLEUW, VMULESW, VMULOUW,
VMULOSW): Add definitions.

gcc/testsuite/ChangeLog:

2017-07-17  Carl Love  <cel@us.ibm.com>

Revert commit r249572  2017-06-22 Carl Love  <cel@us.ibm.com>
test case changes for commit 249424

* gcc.target/powerpc/builtins-2.c (vmulosh, vmulouh, vmulesh,
vmuleuh): Fix scan-assembler-times should check for word not half word
instructions.

From-SVN: r250295

Daily bump.

From-SVN: r250294

alpha.c: Include predict.h.

* config/alpha/alpha.c: Include predict.h.

From-SVN: r250290

re PR tree-optimization/81162 (UBSAN switch triggers incorrect optimization in SLSR)

2017-07-17  Bill Schmidt  <wschmidt@linux.vnet.ibm.com>

PR tree-optimization/81162
* gcc.dg/pr81162.c: Move this to...
* gcc.dg/ubsan/pr81162.c: ...here.

From-SVN: r250284

parser.c (cp_parser_decl_specifier_seq): Add fix-it hints for friend outside class and obsolete auto as...

        * parser.c (cp_parser_decl_specifier_seq): Add fix-it hints for
        friend outside class and obsolete auto as storage-class-specifier.

        * g++.dg/diagnostic/friend1.C: New test.
        * g++.dg/cpp0x/auto1.C: Add check for fix-it hint.

From-SVN: r250282

class.c (maybe_warn_about_overly_private_class): Ignore public copy ctors.

* class.c (maybe_warn_about_overly_private_class): Ignore public
copy ctors.

* g++.dg/warn/ctor-dtor-privacy-3.C: New.

From-SVN: r250281

class.c (type_has_user_declared_move_constructor, [...]): Combine into ...

* class.c (type_has_user_declared_move_constructor,
type_has_user_declared_move_assign): Combine into ...
(classtype_has_user_move_assign_or_move_ctor_p): ... this new function.
* cp-tree.h (type_has_user_declared_move_constructor,
type_has_user_declared_move_assign): Combine into ...
(classtype_has_user_move_assign_or_move_ctor_p): ... this. Declare.
* method.c (maybe_explain_implicit_delete): Use it.
(lazily_declare_fn): Use it.
* tree.c (type_has_nontrivial_copy_init): Use it.

From-SVN: r250280

gcc-dg.exp: Increase expect's match buffer size.

2017-07-17  Bernd Edlinger  <bernd.edlinger@hotmail.de>

        * lib/gcc-dg.exp: Increase expect's match buffer size.

From-SVN: r250279

tree-vrp.c (compare_assert_loc): Fix comparison function to return predictable results.

2017-07-17  Yury Gribov  <tetra2005@gmail.com>

gcc/
        * tree-vrp.c (compare_assert_loc): Fix comparison function
to return predictable results.

From-SVN: r250278

[ARC] Deprecate mexpand-adddi option.

Emitting subregs in the expand will result in broken code due to LRA handling of them. Issue observed while turning on mlra and mexpand-adddi options using dejagnu test suite. Deprecate this
option.

gcc/
2017-04-26  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (adddi3): Remove support for mexpand-adddi
option.
(subdi3): Likewise.
* config/arc/arc.opt (mexpand-adddi): Deprecate it.
* doc/invoke.texi (mexpand-adddi): Update text.

From-SVN: r250276

[ARC] [LRA] Avoid emitting COND_EXEC during expand.

Emmitting COND_EXEC rtxes during expand does introduces errors due to LRA handling of them. Issue discovered while running dejagnu test suit with mlra option on.

gcc/
2017-07-17  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.md (clzsi2): Expand to an arc_clzsi2 instruction
that also clobbers the CC register. The old expand code is moved
to ...
(*arc_clzsi2): ... here.
(ctzsi2): Expand to an arc_ctzsi2 instruction that also clobbers
the CC register. The old expand code is moved to ...
(arc_ctzsi2): ... here.

From-SVN: r250275

[ARC] Enable indexed loads for elf targers.

Enable indexed loads only for elf target, as the linux ones need more testing.

gcc/
2017-02-28  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc.opt (mindexed-loads): Use initial value
TARGET_INDEXED_LOADS_DEFAULT.
(mauto-modify-reg): Use initial value
TARGET_AUTO_MODIFY_REG_DEFAULT.
* config/arc/elf.h (TARGET_INDEXED_LOADS_DEFAULT): Define.
(TARGET_AUTO_MODIFY_REG_DEFAULT): Likewise.
* config/arc/linux.h (TARGET_INDEXED_LOADS_DEFAULT): Define.
(TARGET_AUTO_MODIFY_REG_DEFAULT): Likewise.

From-SVN: r250274

semantics.c (classtype_has_nothrow_assign_or_copy_p): Clarify semantics, simplify implementation.

* semantics.c (classtype_has_nothrow_assign_or_copy_p): Clarify
semantics, simplify implementation.

From-SVN: r250272

Do not allow -fgnu-tm w/ -fsanitize={kernel-,}address (PR sanitizer/81302).

2017-07-17  Martin Liska  <mliska@suse.cz>

PR sanitizer/81302
* opts.c (finish_options): Do not allow -fgnu-tm
w/ -fsanitize={kernel-,}address.  Say sorry.

From-SVN: r250271

re PR tree-optimization/81369 (ICE in generate_code_for_partition)

PR target/81369
* tree-loop-distribution.c (classify_partition): Only assert on
numer of iterations.
(merge_dep_scc_partitions): Delete prameter.  Update function call.
(distribute_loop): Remove code handling loop with unknown niters.
(pass_loop_distribution::execute): Skip loop with unknown niters.

From-SVN: r250270

re PR tree-optimization/81369 (ICE in generate_code_for_partition)

PR target/81369
* tree-loop-distribution.c (merge_dep_scc_partitions): Sink call to
function sort_partitions_by_post_order.

gcc/testsuite
* gcc.dg/tree-ssa/pr81369.c: New.

From-SVN: r250269

re PR tree-optimization/81374 (ICE in bb_top_order_cmp, at tree-loop-distribution.c:391)

PR tree-optimization/81374
* tree-loop-distribution.c (pass_loop_distribution::execute): Record
the max index of basic blocks, rather than number of basic blocks.

From-SVN: r250268

[ARC] Consolidate PIC implementation.

This patch refactors a number of functions and compiler hooks into using a
single function which checks if a rtx is suited for pic or not. Removed
functions are arc_legitimate_pc_offset_p and arc_legitimate_pic_operand_p
beeing replaced by calls to arc_legitimate_pic_addr_p. Thus we have an
unitary way of checking a rtx beeing pic.

gcc/
2017-07-17  Claudiu Zissulescu  <claziss@synopsys.com>

* config/arc/arc-protos.h (arc_legitimate_pc_offset_p): Remove
proto.
(arc_legitimate_pic_operand_p): Likewise.
* config/arc/arc.c (arc_legitimate_pic_operand_p): Remove
function.
(arc_needs_pcl_p): Likewise.
(arc_legitimate_pc_offset_p): Likewise.
(arc_legitimate_pic_addr_p): Remove LABEL_REF case, as this
function is also used in constrains.md.
(arc_legitimate_constant_p): Use arc_legitimate_pic_addr_p to
validate pic constants. Handle CONST_INT, CONST_DOUBLE, MINUS and
PLUS.  Only return true/false in known cases, otherwise assert.
(arc_legitimate_address_p): Remove arc_legitimate_pic_addr_p as it
is already called in arc_legitimate_constant_p.
* config/arc/arc.h (CONSTANT_ADDRESS_P): Consider also LABEL for
pic addresses.
(LEGITIMATE_PIC_OPERAND_P): Use
arc_raw_symbolic_reference_mentioned_p function.
* config/arc/constraints.md (Cpc): Use arc_legitimate_pic_addr_p
function.
(Cal): Likewise.
(C32): Likewise.

gcc/testsuite
2017-07-17  Claudiu Zissulescu  <claziss@synopsys.com>

* gcc.target/arc/pr9000674901.c: New file.
* gcc.target/arc/pic-1.c: Likewise.
* gcc.target/arc/pr9001191897.c: Likewise.

From-SVN: r250267

[PATCH] [ARC] Add support for naked functions.

gcc/
2017-07-17  Claudiu Zissulescu  <claziss@synopsys.com>
            Andrew Burgess  <andrew.burgess@embecosm.com>

        * config/arc/arc-protos.h (arc_compute_function_type): Change prototype.
        (arc_return_address_register): New function.
        * config/arc/arc.c (arc_handle_fndecl_attribute): New function.
        (arc_handle_fndecl_attribute): Add naked attribute.
        (TARGET_ALLOCATE_STACK_SLOTS_FOR_ARGS): Define.
        (TARGET_WARN_FUNC_RETURN): Likewise.
        (arc_allocate_stack_slots_for_args): New function.
        (arc_warn_func_return): Likewise.
        (machine_function): Change type fn_type.
        (arc_compute_function_type): Consider new naked function type,
        change function return type.
        (arc_must_save_register): Adapt to handle new
        arc_compute_function_type's return type.
        (arc_expand_prologue): Likewise.
        (arc_expand_epilogue): Likewise.
        (arc_return_address_regs): Delete.
        (arc_return_address_register): New function.
        (arc_epilogue_uses): Use above function.
        * config/arc/arc.h (arc_return_address_regs): Delete prototype.
        (arc_function_type): Change encoding, add naked type.
        (ARC_INTERRUPT_P): Change to handle the new encoding.
        (ARC_FAST_INTERRUPT_P): Likewise.
        (ARC_NORMAL_P): Define.
        (ARC_NAKED_P): Likewise.
        (arc_compute_function_type): Delete prototype.
        * config/arc/arc.md (in_ret_delay_slot): Use
        arc_return_address_register function.
        (simple_return): Likewise.
        (p_return_i): Likewise.

gcc/testsuite
2017-07-17  Claudiu Zissulescu  <claziss@synopsys.com>
            Andrew Burgess  <andrew.burgess@embecosm.com>

        * gcc.target/arc/naked-1.c: New file.
        * gcc.target/arc/naked-2.c: Likewise.

Co-Authored-By: Andrew Burgess <andrew.burgess@embecosm.com>
From-SVN: r250266

re PR tree-optimization/81428 (ICE: in build_one_cst, at tree.c:2079 with -O2. Fixed point division.)

PR tree-optimization/81428
* match.pd (X / X -> one): Don't optimize _Fract divisions, as 1
can't be built for those types.

* gcc.dg/fixed-point/pr81428.c: New test.

From-SVN: r250265

Remove stuff dead since r239246.

* config/avr/avr-arch.h (avr_inform_devices): Remove dead proto.
* config/avr/avr-devices.c (mcu_name, comparator, avr_mcus_str)
(avr_inform_devices): Remove dead stuff.

From-SVN: r250264

gcc_update (files_and_dependencies): Remove dead entry.

* gcc_update (files_and_dependencies)
[gcc/config/avr/avr-tables.opt]: Remove dead entry.

From-SVN: r250263

arm_neon.h: Fix softp typo.

2017-07-17  Tamar Christina  <tamar.christina@arm.com>

* config/arm/arm_neon.h: Fix softp typo.

From-SVN: r250262

re PR tree-optimization/81365 (GCC miscompiles swap)

PR tree-optimization/81365
* tree-ssa-phiprop.c (propagate_with_phi): When considering hoisting
aggregate moves onto bb predecessor edges, make sure there are no
loads that could alias the lhs in between the start of bb and the
loads from *phi.

* g++.dg/torture/pr81365.C: New test.

From-SVN: r250261

re PR middle-end/80929 (Division with constant no more optimized to mult highpart)

PR 80929
* config/avr/avr.c (avr_mul_highpart_cost): New static function.
(avr_rtx_costs_1) [TRUNCATE]: Use it to compute mul_highpart cost.
[LSHIFTRT, outer_code = TRUNCATE]: Same.

From-SVN: r250258

re PR tree-optimization/81396 (Optimization of reading Little-Endian 64-bit number with portable code has a regression)

PR tree-optimization/81396
* tree-ssa-math-opts.c (struct symbolic_number): Add n_ops field.
(init_symbolic_number): Initialize it to 1.
(perform_symbolic_merge): Add n_ops from both operands into the new
n_ops.
(find_bswap_or_nop): Don't consider n->n == cmpnop computations
without base_addr as useless if they need more than one operation.
(bswap_replace): Handle !bswap case for NULL base_addr.

* gcc.dg/tree-ssa/pr81396.c: New test.

From-SVN: r250257

Insert diverging jump alap in nvptx_single

2017-07-17  Tom de Vries  <tom@codesourcery.com>

PR target/81069
* config/nvptx/nvptx.c (nvptx_single): Insert diverging branch as late
as possible.

From-SVN: r250256

[SPARC/RTEMS] Add __FIX_LEON3FT_B2BST

In case the LEON3FT back-to-back store workaround is active
(sparc_fix_b2bst), then define the builtin define __FIX_LEON3FT_B2BST on
RTEMS.  The intended use case for this is operating system code in
assembly language.  See also:

https://lists.rtems.org/pipermail/devel/2017-July/018463.html

gcc/
*  gcc/config/sparc/rtemself.h (TARGET_OS_CPP_BUILTINS): Add
conditional builtin define __FIX_LEON3FT_B2BST.

From-SVN: r250254

[RTEMS] Add multilibs for LEON3FT back-to-back store workaround

Replace MULTILIB_EXCEPTIONS with MULTILIB_REQUIRED for readability.
-mfix-gr712rc and -mfix-ut700 are currently equivalent.

gcc/
* config/sparc/t-rtems: Add mfix-gr712rc multilibs. Replace
MULTILIB_EXCEPTIONS with MULTILIB_REQUIRED. Match -mfix-gr712rc
with -mfix-ut700.

From-SVN: r250252

Daily bump.

From-SVN: r250251

re PR rtl-optimization/81424 (internal error on GPRbuild with -O2)

PR rtl-optimization/81424
* optabs.c (prepare_cmp_insn): Use copy_to_reg instead of force_reg
to remove potential trapping from operands if -fnon-call-exceptions.

From-SVN: r250246

tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Use profile_proability for scalling.

* tree-ssa-loop-manip.c (tree_transform_and_unroll_loop): Use
profile_proability for scalling.
* scale_profile_for_vect_loop.c (scale_profile_for_vect_loop): Likewise.

From-SVN: r250245

* cgraph.c (cgraph_edge::redirect_call_stmt_to_caller): Cleanup.

From-SVN: r250243

cfgloopmanip.c (scale_loop_profile): Avoid use of REG_BR_PROB_BASE fixpoint arithmetics.

* cfgloopmanip.c (scale_loop_profile): Avoid use of REG_BR_PROB_BASE
fixpoint arithmetics.

From-SVN: r250242